qlnxe: Avoid reinitializing the interface when it is already initialized

[zlei]

Summary

qlnx_init_locked() unconditionally uninitialize the interface thus is
actually reinitializing the interface. Well the init routine qlnx_init()
is to initialize the interface by net stack when assigned with the first
inet or inet6 address. The ioctl SIOCSIFADDR for the first inet6 address
is handled by ether_ioctl() thus the interface is reinitialized no matter
it was initialized or not.

Add a driver status check for that to avoid reinitializing. Further plan
is removing SIOCSIFADDR ioctl from the driver and let ether_ioctl() handle
it.

MFC after: 1 week

Test Plan

1. Assign an IPv4 address to ql0, or bring it up via ifconfig ql0 up.
2. Assign the first IPv6 address to ql0.

Watch /var/log/messages, there should be no ql0: link state changed to DOWN event while assigning the IPv6 address.

[kbowling]

Light ack as the two init functions seem fine but I have not looked through the entirety to see if there was some reason for stopping the MAC (rdma?) and you should probably look at all the leafs of qlnx_load especially filter setup.

[zlei]

In D54887#1260593, @kbowling wrote:

Light ack as the two init functions seem fine but I have not looked through the entirety to see if there was some reason for stopping the MAC (rdma?) and you should probably look at all the leafs of qlnx_load especially filter setup.

The function qlnx_slowpath_start() is due to enable iWARP or RoCE. It is called on attaching the device ( port ), or error recover, or manually triggered by sysctl qlnx_trigger_dump_sysctl() [1], but not by qlnx_load() . The latter two will also do qlnx_stop() and qlnx_load() to recover the device.

static int
qlnx_slowpath_start(qlnx_host_t *ha)
{
        struct ecore_dev        *cdev;
        struct ecore_pf_params  pf_params;
        int                     rc;

        memset(&pf_params, 0, sizeof(struct ecore_pf_params));
        pf_params.eth_pf_params.num_cons  =
                (ha->num_rss) * (ha->num_tc + 1);

#ifdef QLNX_ENABLE_IWARP
        if (qlnx_vf_device(ha) != 0) {
                if(ha->personality == ECORE_PCI_ETH_IWARP) {
                        device_printf(ha->pci_dev, "setting parameters required by iWARP dev\n");       
                        pf_params.rdma_pf_params.num_qps = 1024;
                        pf_params.rdma_pf_params.num_srqs = 1024;
                        pf_params.rdma_pf_params.gl_pi = ECORE_ROCE_PROTOCOL_INDEX;
                        pf_params.rdma_pf_params.rdma_protocol = ECORE_RDMA_PROTOCOL_IWARP;
                } else if(ha->personality == ECORE_PCI_ETH_ROCE) {
                        device_printf(ha->pci_dev, "setting parameters required by RoCE dev\n");        
                        pf_params.rdma_pf_params.num_qps = 8192;
                        pf_params.rdma_pf_params.num_srqs = 8192;
                        //pf_params.rdma_pf_params.min_dpis = 0;
                        pf_params.rdma_pf_params.min_dpis = 8;
                        pf_params.rdma_pf_params.roce_edpm_mode = 0;
                        pf_params.rdma_pf_params.gl_pi = ECORE_ROCE_PROTOCOL_INDEX;
                        pf_params.rdma_pf_params.rdma_protocol = ECORE_RDMA_PROTOCOL_ROCE;
                }
        }
#endif /* #ifdef QLNX_ENABLE_IWARP */

        cdev = &ha->cdev;
...
}

If there is some reason for stopping the MAC, a pair of SIOCSIFFLAGS ioctls can be issued, say ifconfig ql0 down && ifconfig ql0 up.

[1] Well I'm hacking the qlnx_slowpath_start(), I got kernel panics. From the stack trace, it appears some locks are not freed properly. I'm still investigating on it but I think that is not urgent.